Some kind of comparison between different SRC:s (including Weiss new software SRC). R8brain is participating.
Have you seen it Aleksey? Any comments?
This test does not include the latest 'ultra-steep' mode of r8brain PRO. Otherwise, this test seems to be objective. r8brain PRO, with its 150 dB (for r8brain PRO) signal-to-noise ratio for individual spectral bins is pretty good to my knowledge.
What this test is lacking of is: aliasing test. I have optimized r8brain PRO to give no aliasing at all. I suspect Weiss Saracon (and, quite strangely, iZotope SRC) start to alias too EARLY: at -7 dB around 22kHz, with r8brain PRO that's -150 dB at 22kHz. In Ultra-steep mode that's the same - only passband is considerably wider.
Another interesting test could be conversion time.
For the sake of numbers only I guess I'll need to go 64-bit resolution somewhere in the future. This should give up to 330 dB signal-to-noise ratio, but that's of course, a total overkill. When converting that to 32-bit floats, we'll have a lower signal-to-noise ratio.
But do not forget about sound. If most of these sample rate converters are linear-phase, they can't compare 1-to-1 to minimum-phase mode of r8brain PRO sound wise.
What makes me wonder is how they have peformed these measures. When you convert 96kHz to 44.1kHz your highest resulting frequency is 22.05kHz. But they have 2 Megaherz on their graphs. Too strange.
OK, we'll have to listen to responses from the actual users, and compare prices as well. Other than not-so-exagerrated SNR, I do not see problems with r8brain PRO at the moment.
I do not think dithering gives as much difference as SRC algorithm itself. Moreover, due to the nature of sample rate conversion (it may generate over 0 dBFS overshoots) I usually suggest to convert to 32-bit float depth (no dithering takes place) and then use waveeditor to first process the converted track with a peak-limiter and then convert it to a lower bit-depth using any favorite bit-depth reduction algorithm.
In this scenario, bit-depth conversion does not take a part in the process at all.
The SRC shootout website has been updated to include the latest R8brain Pro 1.3.1 with Ultra Steep, both in linear and minimum phase modes.
Copying my answer at the Nuendo Lounge forum:
"They've updated their test to include r8brain PRO 1.3.1 - that's cool... But their measurements are not precise (awkward I would say). I do not know why they've got a slight fall-off in min.phase mode before it goes steep. According to my measurements (based on direct filter evaluation), frequency response of lin.phase and min.phase modes is equal in the transition band - bit-to-bit identical up to 3rd decimal point (in dB scale).
And I still don't get their 2 MHz measurements. r8brain PRO does not produce such high frequencies. If that's extrapolation, it has nothing to do with the converter itself.
BTW, Saracon should tighten their transition response now. :) They win in the SNR department, but now obviously lose in transition response. :)"
This topic was last updated 180 days ago, and thus it was archived. Replying is disabled for this topic.